• ## Summary

Search form
• 00:01

[MUSIC PLAYING][RESEARCH METHODS][An Introduction to Errors in Sampling]

• 00:11

CHIRANJEEV KOHLI: I'm Chiranjeev Kohli, Professor of Marketingat California State University of Fullerton.[Chiranjeev Kohli, PhD, Professor of Marketing,California State University, Fullerton]I've taught marketing research, marketing strategy,and advertising.And I've done extensive consulting, mostlywith Fortune 1000 companies.My core areas of expertise are branding, branding research,and pricing research.

• 00:34

CHIRANJEEV KOHLI [continued]: The topic for this module is errors and sampling.Before we even get started, it's important to understandthe purpose of sampling.While we do sampling in all kinds of research,it becomes particularly important whenwe're dealing with larger populations,especially when doing quantitative survey research.

• 00:56

CHIRANJEEV KOHLI [continued]: We do sampling when it is difficult, expensive,or impossible to measure everyone in the population.That's when we rely on a smaller group, or a sample,to learn about the population.Whenever we are sampling, by definition,we are trying to estimate the population value.

• 01:17

CHIRANJEEV KOHLI [continued]: Population here implies the census of all the elementswe are interested in.This could be the population of a county,or all the students in grade 8, or all the millionairesin a certain area--any group we are interested in learning about.The word "estimate" is also key here,because we may not know the population value.

• 01:38

CHIRANJEEV KOHLI [continued]: If we know the population value, thereis no need for sampling in the first place.So any time we are doing sampling,our hope is to come up with an unbiased estimateof the population value.So we are after this elusive population value,which may also be referred to as the true valuefor the population.

• 02:02

CHIRANJEEV KOHLI [continued]: The sample value is only an estimateof the population value.When we are sampling, we will neverknow what the population value is,but our hope is that the estimate we get from the sampleis as close to the population value as possible.This is easier said than done, because errors creep upin at several levels.

• 02:26

CHIRANJEEV KOHLI [continued]: Let's talk about these errors.Let's use the example of a hypothetical personal securitydevice to explain sampling errors.Say you wear something on your wrist,and if you're in harm's way, you press a button on itwhich alerts the police.What we want to know is, what percentage of residentsof a certain county would be interested in purchasing it?

• 02:51

CHIRANJEEV KOHLI [continued]: That's the population value we're interested in.Now, there are four kinds of sampling errorswhich will creep in.Let's discuss them one by one.To get started with sampling, we need a sampling frame.Sampling frame refers to the list from whichwe will be drawing a sample.

• 03:12

CHIRANJEEV KOHLI [continued]: In this case, we may decide to do random digit dialing usingthe area codes for that county.Let's represent the residents of this countywith a black rectangle in this diagram.This is the population of interest.All phone owners constitute the sampling frame in our example,because we're using random digit dialing.

• 03:36

CHIRANJEEV KOHLI [continued]: This is represented by the red rectangle in our diagram.Now, if there is a perfect overlap between the two,we're good.However, that's often not the case.So we identify three different areas--1, 2, and 3 in the diagram.

• 03:59

CHIRANJEEV KOHLI [continued]: Area 1 is the overlap area, so that is not a problem.Area 2 is in the sampling frame, but not in the population.So that is a problem.Respondents in the area 2 are easy to address.We ask a qualifying question.Something like, are you a resident of this county?

• 04:21

CHIRANJEEV KOHLI [continued]: It's possible that they are not residents,but they have a phone number from that area.Maybe because after getting their phone numbers,they had moved out of the county.If so, they would respond with a no to this qualifying question,and we can remove them from our sample.

• 04:44

CHIRANJEEV KOHLI [continued]: Area 3 is in the population, but not in the sampling frame,so that is also a problem.This one gets a little bit tricky.These are people we should have included, but we did not.This leads to the potential for error.In our example, these might be people without phones--say homeless people.

• 05:06

CHIRANJEEV KOHLI [continued]: Now if these people would have responded the same wayas people in area 1, frame error would notbe an actual problem, even though itwas a potential problem.However, if these people had responded differentlythan people in area 1, we have an actual error,because our estimate is based only on area 1,but it is supposed to be an estimatefor the entire population.

• 05:37

CHIRANJEEV KOHLI [continued]: So whatever estimate we get from area 1,we'll have to use some adjustmentto come up with the more accurate estimate reflectiveof the entire population.Broadly speaking, there are two waysof doing this, neither of which is perfect.The first is to do some subjective adjustment.

• 05:57

CHIRANJEEV KOHLI [continued]: The homeless are less likely to purchase the device becauseof economic reasons, so we do a downward revisionof our estimate based on area 1.Another way is to put extra effortinto reaching out to some subjects in area 3through, say, some other approach.

• 06:18

CHIRANJEEV KOHLI [continued]: Find out the value for that group,and use it to adjust our estimate from area 1.Now, with frame error already in place,we discuss non-response error.Remember, we are only reaching out to people in area 1 now.Some of these people will respond.Those subjects are represented by area 1A.

• 06:42

CHIRANJEEV KOHLI [continued]: Others will not respond, and they're represented by area 1B.Once again, we will be basing our assessmenton the people who responded--that is, people in area 1A.If subjects in 1B were to respond differentlythan those in 1A, then we have potential for another error.

• 07:06

CHIRANJEEV KOHLI [continued]: This is called the non-response error.To fix this error, we have to figure outhow 1B may have responded differently than 1A.Typically, people who do not respond to a surveyare less interested in what the survey's all about,so it's fair to say that estimate we got from 1Awill be an overestimate of the population value,because 1B was excluded.

• 07:37

CHIRANJEEV KOHLI [continued]: Broadly speaking, there are two wayswe can address this non-response error.We can send out the survey in multiple waves.Subjects who are more interested aremore likely to respond in the earlier waves,so the later waves give us a better indication of whatthe value may be for 1B.

• 07:58

CHIRANJEEV KOHLI [continued]: In our example, people in the area 1Bare less likely to be interested in our product.This inflates our estimate, because it is onlybased on people in 1A.So we use the values from later wavesto lower our estimate based only on 1A.

• 08:20

CHIRANJEEV KOHLI [continued]: Another way of addressing this is to put an extra effortmeasure--a subsample of 1B--and use that to adjust our estimate based only on 1A.Now that frame error and non-response errorare in place, or have been adjusted for hopefully,we move to the third type of error,called the response error.

• 08:44

CHIRANJEEV KOHLI [continued]: This error comes into play becauseof poorly-worded questions.There's a whole list of issues with question wording,but that is discussion for another topic.However, as one example, if we were to just ask people,are you interested in this product?We're likely to get an inflated estimate.That's because people have a tendencyto agree with whatever is being asked in the question.

• 09:08

CHIRANJEEV KOHLI [continued]: However, as one example, if you were to just ask people,are you interested in this product?We are likely to get an inflated estimate.That's because people have a tendencyto agree with whatever is being asked in the question.Technically, these are called implied alternative questions.The solution to this lies in using carefully-worded surveyquestions to avoid this kind of error.

• 09:34

CHIRANJEEV KOHLI [continued]: The last type of error is called random error.This is, by definition, chance error,and not a systematic one.We cannot say if it will lead to an overestimateor an underestimate, because this is random.The solution to this lies in reducing the random error.One way of doing this is by using multi-item scalequestions, which will likely lead to random errorsacross questions compensating each other.

• 10:01

CHIRANJEEV KOHLI [continued]: Another way of addressing this is at the sample level--is to use large sample sizes.The larger the sample size, the smaller the random error.This will result in overestimation,or an underestimation, but the size of the errorcan be calculated and reduced, even though it is onlyin a probabilistic manner.

• 10:26

CHIRANJEEV KOHLI [continued]: But it can never be completely eliminated.There you have it--the four types of sampling errors, and how to fix them.

### Video Info

Publisher: SAGE Publications Ltd.

Publication Year: 2020

Video Type:Tutorial

Methods: Sampling error, Sampling, Marketing research

### Segment Info

Segment Num.: 1

Persons Discussed:

Events Discussed:

Keywords:

## Abstract

Chiranjeev Kohli, PhD, Professor of Marketing at California State University, Fullerton, discusses four common errors in sampling—frame, non-response, response, and random— and ways to address them.

## Looks like you do not have access to this content.

An Introduction to Errors in Sampling

Chiranjeev Kohli, PhD, Professor of Marketing at California State University, Fullerton, discusses four common errors in sampling—frame, non-response, response, and random— and ways to address them.

Copy and paste the following HTML into your website